Detection of Bilateral Tonic-Clonic Seizures Using Miniaturized Wearable Electromyography-Accelerometry Sensors.

Abstract

This study aimed to develop a seizure detection algorithm using surface electromyography (sEMG) and accelerometry (ACC) signals recorded with miniaturized wearable sensors. Continuous sEMG-ACC signals were acquired from patients wearing eight sensors positioned bilaterally on the upper trapezius, anterior deltoid, biceps brachii, and tibialis anterior muscles. We trained an extreme gradient boosting classifier to identify seizure epochs using setups with eight, two, and one sensor(s). Performance was evaluated via patient-wise nested cross-validation, and specificity was further assessed on an independent patient cohort without seizures. Eleven generalized tonic-clonic seizures (GTCS) and focal-to-bilateral tonic-clonic seizures (FBTCS) were recorded from nine patients over 1359.6 h. The best results were obtained with a dual-sensor setup combining data from the right biceps brachii and the left tibialis anterior, achieving 100% sensitivity, 0.12 FAR/24h, and median detection latency of 22 s. On 1744.18 h of data from 19 patients without seizures, FAR/24h was 0.06. The developed algorithm effectively detected GTCS and FBTCS in an epilepsy monitoring unit, even with a reduced number of sensors. This approach could enable timely interventions in outpatient settings, potentially improving safety and independence for people with epilepsy.